Collaborative Filtering Based Recommender System Research And Its Applications

E-commerce is developing rapidly during the period of the mobile Internet. The number of transactions between customers and businesses is also increasing multiply,which resulted in the situation of "information overload". Although this technology has a more excellent performance in the Internet platform, but there are some unsolved problems. The extreme sparsely of item-space is the most important problem in the study of collaborative filtering technology in recent years.With the development of mobile Internet technology, more and more customers and goods accumulate in data base. Due to the record of customer's score for the goods is limited, so the score of each item will be relatively reduced in e-commerce database, which result in the data of sparse. In this thesis, we start with the problem of sparse data, and improving the traditional collaborative filtering algorithm. The main researching contents can be described as follows:(1)In order to improve the problem of high dimension and the cold-start of users,this thesis proposed a collaborative filtering recommendation algorithm based on weighted prediction of double directional clustering. The algorithm is clustering the users and items based on their similarity. Then the clustering data is considered as the initial data, which can generate recommendations for the target users. When someone needs to be recommended, the clustering center of each data cluster is calculated to be similar to it, and attributing it to the data cluster of the highest similarity to recommend. In the end, combine both the prediction of user and item to generate recommendation. The experimental results show that the algorithm can reduce the dimension of project space, reduce the sparsity of data, enhance the reliability of similarity between each user, and it also improves the user's cold start problem.(2)In order to improve the problem of low score-density, which can result in inaccurate the similarity of each user, this thesis proposed a collaborative filtering algorithm based on predicting and filling miss-data by iterating. The algorithm is to improve the sparsity of data and the density of score. The algorithm uses the similarity of the item to predict the score and fill the missing values in the score matrix. At the same time, the score-matrix is filled again and again in by an iterative method until the density of scored converges to a certain value. It can improve the accuracy of the calculation of the similarity between each user(or items).The experimental results show that the algorithm can effectively improve the density of scored and the calculation error of similarity.(3)This thesis will use Movielens dataset to validate the innovative content of this thesis. The experimental results show that the proposed algorithm of this thesiscan be used to improve the user's recommendation accuracy with the high dimensional data set.